To: David K. who wrote (13157 ) 6/19/1999 2:32:00 AM From: Ben Wu Read Replies (1) | Respond to of 14577
hi y'all, The 3dfx/S3 rumor has caused me to jump boards for a while and see what's going on on the S3 side. i'll try to take a pot shot at trying to explain the article. There are two qualities to a 3d card that define it's performance. fillrate and image quality. Fill rate describes how fast a card can render a given screen and apply textures to it, this directly affects the frame rate of a given card. Image quality is more general/qualitative measure as it's name suggests. It's a combination of color depth (32 bit vs. 16 bit) and how well a card can render a given screen (i.e. if fog looks correctly, or if the sky looks screwed up, low gamma values, washed out colors, etc. etc.) A TMU is a texture mangement unit. What it does is it renders a texture onto a surface. Why is it important? because it allows for all those cool surfaces to applied to the polgon skeleton. Starting from Voodoo 2, 3d cards began to add a "second" TMU to the chipset. Why is a second TMU important? because starting from Quake 2, 3d engines began using multiple textures to render a surface. This allows for rendering of the initial surface and adding another texture on top of it to give it a blended more realistic look. Graphics cards that only have one TMU (eg 3dfx Banshee) found themselves at a huge disadvantage when playing games like quake2, ie. they take twice as long to render a frame, thus halving the frame rate. So when a card is refered to having single-pass-multi-texturing, this generally refers to it's ability to render two textures simultaneously. All current generation cards do this. However, as 3d engines evolve, cards may be required to handle 3, 4, or even more textures per triangle. Next generation cards need to be able to handle these requirement, so the more TMUs the better. 3dfx and nVidia are both rumored to have 4+ TMUs. Geometry acceleration is another feature of the chip. What geometry acceleration does is it handles all the point setups and triangle calculations for a given scene. Currently the CPU handles that, and needless to say, it's a HUGE performance hit. That is why the cards of today are often quoted as being "CPU-limited". a card can only render as fast as the rate it receives information from the CPU. What geometry acceleration does is it helps calculate the trianges for rendering instead of waiting for the CPU to calculate the triangles. This is the next generational leap for 3D processors. Essentially, this will greatly eliminate the CPU overhead, thus making the CPU bottle neck a thing of the past. 3dfx's "napalm" and nVidia's "NV10" (code name for their next gen parts) both are rumored to have geometry acceleration. Transform and Lighting. What transform and light is generally used for is calculating the "shinyness" of a tile. For example, when you fire a rocket in Quake 2, as it passes down the hall, you will be able to see the light from the rocket's flame on the walls. This is also very CPU intensive. nVidia has publically announced that they will be the FIRST chip company to have T&L transforms on chip. But recent rumors suggest that nVidia may be blowing smoke again (nVidia is notorious for making promises it cannot keep). We may not see T&L transforms on a chip in this upcoming generation (there are various developer issues involved). Memory bandwidth (64-bit vs. 128 bit). This is pretty much a no-brainer. The more bits the CPU can access, the better. But 64-bit memory is currently a lot cheaper than 128-bit. sticking with 64-bit is currently a cost issue. Voodoo3 is 128-bit only, TNT2 can handle both, and S4 is 64 bit only. That is why you can see 32 mb S4 boards costing the same as 16 MB V3 boards. I hope this helps. I know i probably missed something somewhere, but this is the best summary i can come up with off the top of my head. Okay, now back to investment discussion. What concerns me about S3 as a long term investment is it's ability to stay in black. Talk among this board currently focuses on S3's investments in the various foundries. But, if i'm not mistaken, those same contracts lock S3 into buying certain amounts of chips from the foundries despite whether they can sell them or not (think NSM when it first bought out CYRIX and the IBM deal). At this point, it looks to me like S3 appears to be selling it's chips very near cost in order to move its chips and regain market share. In the arena that S3 has chosen to compete in, it has VERY formidable competition (ATI Rage128, nVidia Vanta, and Intel i752 chips). Without a high performance chip, S3's margins will stay low (think AMD vs. Intel here). Unless a miracle happens, i don't see S3 taking the performance lead anytime soon from 3dfx and nVidia. Honestly, i don't expect much from the GX4. However, the .18u process may reduce chip costs enough to increase the margins on the them. I'm not here to bash S3. like a lot of people, i'm just trying to sort out the issues that S3 has facing it. Plus the 3dfx/S3 rumors have rustled up some curiosity and interest on my part. I find the rumors very intriguing to say the least. S3 messed up royally by missing the 3d boat and it has been playing catch up ever since (it's turnaround in the past year has been pretty impressive however). Right now S3 has a part that will allow it to compete in the low end. For how long? The next quarterly report should provide a rough picture. -ben btw, the 3dfx board is generally full of information regarding 3d technology and trends in general, so you might want to listen in periodically.